Washington: From an obscure experiment in 2017, the image-processing AI known as Maven has grown into one of the military’s hottest tools for intelligence and targeting, with plans to expand its user base ten-fold. But that growth poses a logistical problem for the National Geospatial Intelligence Agency, which took responsibility for running Maven as a formal Pentagon program of record last year.
According to the three-star director of the NGA, Vice Admiral Frank Whitworth, that rising demand for Maven — and Maven’s laudable flexibility in letting users build custom code and even their own AI models — has begun to strain his agency’s resources. Or as he put it, “Success begets challenges.”
As a general rule, text requires less data than imagery, which requires less than video — and Maven was originally built to detect anomalies in full-motion video from Predator drones, whose sheer volume was drowning the military’s cadre of intelligence analysts.
It was “thousands of hours” of full motion video, Bob Work, who launched Maven in his days as Deputy Secretary of Defence, said during the CSET conference. “We would have seven-person teams that worked seven days a week, and they could essentially process 15 percent of the take and the other 85 percent just was thrown out. [So] let’s use machine learning on the Preds to look at all the take, 100 percent.”
As it’s evolved, the Maven AI not only helps the humans by figuring out which intelligence imagery and video is worth their looking at: It also provides a shared digital repository of information that any user — regular or casual — can access at their desk. So no wonder Maven became so popular with harried staffs struggling to find the needles of actionable intelligence in the ever-growing haystack of digital data.
